Machine learning models to evaluate mortality in pediatric patients with pneumonia in the intensive care unit

Area of Science:

• Pediatric critical care medicine
• Machine learning in healthcare
• Predictive analytics in medicine

Background:

• Pneumonia is a leading cause of mortality in children admitted to the intensive care unit (ICU).
• Accurate prediction of mortality is crucial for timely intervention and resource allocation.
• Existing prediction tools may not fully leverage the potential of machine learning for complex pediatric cases.

Purpose of the Study:

• To develop and validate machine learning models for predicting mortality in pediatric patients with pneumonia admitted to the ICU.
• To identify key clinical features that contribute to mortality prediction.
• To support clinical decision-making in managing critically ill children with pneumonia.

Main Methods:

• Retrospective cohort study including 1231 pediatric ICU admissions for pneumonia (2010-2019).
• Development of two tree-structured machine learning models to predict ICU mortality and 24-hour ICU mortality.
• Utilized 33 features from electronic health records, including demographics, comorbidities, vital signs, and laboratory data.

Main Results:

• The models achieved high predictive performance, with Area Under the Receiver Operating Characteristic Curves (AUROCs) of 0.80 for ICU mortality and 0.92 for 24-hour ICU mortality.
• Key predictors of increased mortality included reduced blood pressure, decreased peripheral capillary oxygen saturation (SpO2), and elevated partial pressure of carbon dioxide (PCO2).

Conclusions:

• Machine learning models demonstrate significant potential in predicting ICU mortality for children with pneumonia.
• These predictive tools can aid clinicians in decision-making, particularly in resource-limited environments.
• Further validation and implementation of these models could improve outcomes for critically ill children.